Kadrey v. Meta: what did the fair use ruling decide, and what did it leave open?

What did the court decide in Kadrey v. Meta?

The short answer: in June 2025 a federal district court in California ruled for Meta on fair use in a case brought by a group of authors, on a narrow basis. As publicly reported, the ruling rested on the plaintiffs' failure to develop evidence of market harm, and the court stressed that it did not establish that training AI models on copyrighted works is lawful in general.

This is a summary of publicly reported information about one trial-court decision. We have not linked the opinion here, so read it, or ask counsel to, before relying on any detail. It is not a substitute for the opinion itself. This is general information, not legal, tax or financial advice. Confirm with your own counsel before acting.

What is the case about?

Kadrey v. Meta Platforms is a copyright suit in the Northern District of California. A group of authors alleged that Meta copied their books to train its Llama language models. Meta argued that the training was fair use under the Copyright Act. In June 2025 the court granted summary judgment to Meta on that defense as to the plaintiffs before it.

Fair use is a case-specific defense. Courts weigh four factors: the purpose and character of the use, the nature of the work, the amount used, and the effect on the potential market for the original. In this case the market-effect factor carried the decision.

Why did market harm matter so much?

Because, as reported, the court treated it as the decisive factor. Coverage describes the court as finding that the plaintiffs had not shown Meta's copying harmed the market for their books, and as noting that a different record could produce a different result.

The opinion also discusses arguments about a market for licensing works as training data. Commentators read that passage differently, and it is the part most relevant to data licensing. Read it in the opinion itself rather than relying on summaries, ours included, and expect it to be debated in later cases.

What does the ruling not decide?

Other AI copyright cases are pending or decided, each on its own record. Compare opinions on their own facts rather than treating any one as a rule.

How do these questions apply to common partner situations?

What does this mean for company data specifically?

The case concerned published books, not internal business records. Company records are different: the company usually owns the rights in documents its employees wrote within their jobs, and its data is held under confidentiality rather than published. The question for a licensing deal is less "is training fair use?" and more "does the company have the right to license this data, and what do customers' contracts and privacy promises say?"

For that reason, rights review sits at the front of the process. The guide on why metadata raises data value covers documentation buyers look for, and the guide to enterprise AI data licensing deals shows how licensing works in practice. For a comparison of regimes, see the UK debate on AI training and copyright.

The US Copyright Office has also published a multi-part report on copyright and AI, including a part on generative AI training released as a pre-publication version in May 2025. It is a report, not law; the Copyright Office AI page lists the parts.

What should a partner say?

Keep it factual and avoid legal conclusions.

How do partner rewards work?

Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 per referred company, payable only after the buyer pays and SourceX receives its fee. No reward is guaranteed.

Questions to ask counsel

• Does this ruling change how we should think about licensing our records?
• Which of our records contain third-party content we cannot license?
• What warranties would we be comfortable giving a buyer about ownership?
• Which court decisions and statutes are relevant to our data type?

What should an owner take from the ruling?

Three practical points follow, none of them legal conclusions.

1. Do not treat one ruling as settled law. It is a trial-court decision on a specific record, and as reported the court itself warned against reading it broadly.
2. Expect the debate to continue. Courts, the Copyright Office and legislatures are all working on how copyright applies to training. Outcomes can differ by facts and by jurisdiction.
3. Clear permission is still the cleaner route. An owner who licenses records through a signed agreement knows the scope, term and price rather than relying on a defense argued after the fact.

Owners sometimes ask whether this means the licensing market is shrinking. The honest answer is that nobody can say from one decision. Business records differ from published books in kind: they are confidential, held by the company and not available to copy from the public web in the first place. Buyers who want them need the company's cooperation, and a signed license is how that cooperation is documented. Rights review, including checking that the company owns what it offers, comes first in any SourceX process.

Next step

If you know a company that holds years of operational records and wants to license them, register as a partner. Review how it works and run a preliminary screen with the company fit checker.